👋Hi!

I am a second-year Ph.D. candidate supervised by Prof. Xin Zhang at Programming Languages Lab, Software Engineering Institute, School of Computer Science, Peking University.

I received my B.Eng. in Software Engineering from Nanjing University in 2024. During my undergraduate studies, I was fortunate to work with the Pascal Research Group under the supervision of Prof. Yue Li and Prof. Tian Tan, where I conducted research on improving the compilation efficiency of foundational static-analysis frameworks for languages running on the Java Virtual Machine.

My research interests lie in AI infra (especially machine learning compiler), program analysis, program synthesis, machine learning, and their intersections. More broadly, I am interested in code- and reasoning-related tasks, especially program comprehension, verification, and efficient inference in the era of large language models.

📝Publications

• 🍺Beer: Interactive Alarm Resolution in Bayesian Program Analysis via Exploration-Exploitation
  Haoran Lin, Zhenyu Yan, Xin Zhang
  TL;DR: We incorporated the exploration-exploitation scheme in Bayesian program analysis to improve the learning effectiveness. The enhancement achieves up to 32% higher ranking efficiency compared to greedy baselines across multiple analysis types.
  ACM SIGPLAN International Conference on Object-Oriented Programming Systems, Languages, and Applications (OOPSLA 2026), October 2026
  [DOI] [Artifact]

• LLM-Based Alarm Resolution Guided by Bayesian Program Analysis
  Yifan Zhang, Yuanfeng Shi, Haoran Lin, Yingfei Xiong, Xin Zhang
  TL;DR: Instead of asking LLMs to directly judge static-analysis alarms—which are difficult to reason about and often share causes and reasoning paths, wasting tokens on redundant reasoning—we have LLMs verify easier-to-check, high-impact intermediate facts. One refuted fact can prune many downstream alarms, avoiding repeated reasoning and token waste. Across 17 Java/C programs, this improves F1 by 2.3-4.4× while using 76.6% fewer tokens.
  ACM SIGPLAN International Conference on Object-Oriented Programming Systems, Languages, and Applications (OOPSLA 2026), October 2026
  [To Appear]

• 🏎Two Approaches to Fast Bytecode Frontend for Static Analysis
  Chenxi Li, Haoran Lin, Tian Tan, Yue Li
  TL;DR: For static analysis, we developed a new Java-bytecode-to-3AC compiler (the frontend of a static-analysis framework) that achieves a 14× speedup across diverse real-world projects compared to the peer compilers in the state-of-the-art frameworks Soot and WALA.
  ACM SIGPLAN International Conference on Object-Oriented Programming Systems, Languages, and Applications (OOPSLA 2025), October 2025
  [DOI] [Artifact]